
Discover more from A Few Things....
A Few Things: Marko on Markets, The Future of Work, Solving American Obesity, Family Office Thematic Investing, LLM 101, There is No AI, How To Capture the AI Moment, A16Z on Crypto.....
May 5 2023
I am sharing this weekly email with you because I count you in the group of people I learn from and enjoy being around.
You can check out last week’s edition here: Ferguson on The USD, What's Happening in PE, Druck on Markets, Where are Middle Eastern $'s Going?, Power and Prediction, What's Next in Gen AI, Charts and News You Might Have Missed....
Believe it or not, that “♡ Like” button is a big deal – it serves as a proxy to new visitors of this publication’s value. If you enjoy this article, don’t be shy :)
Quotes I Am Thinking About:
“The man who will use his skill and constructive imagination to see how much he can give for a dollar, instead of how little he can give for a dollar, is bound to succeed.”
- Henry Ford
“Honesty is the first chapter in the book of wisdom.”
- Thomas Jefferson
“Progress is impossible without change, and those who cannot change their minds cannot change anything.”
- George Bernard Shaw
“To find fault is easy; to do better may be difficult.”
- Plutarch
“The poor wish to be rich, the rich wish to be happy, the single wish to be married, and the married wish to be dead.”
- Ann Landers
A. A Few Things Worth Checking Out:
1. Spoke with my favourite market strategist Marko Papic last week. This was his take:
He thinks we have a US soft-landing, but even if a recession occurs, it will be a strange one, with US consumers flush with deposits, engrossed with swollen home equity, and completely deleveraged. His high conviction view that the Fed will lean towards leniency no matter what the scenario. Bet on leniency ahead of the US election.
The debt ceiling crisis is a definite risk. Investors should “sell in May and go away” as the summer could present a tactical opportunity to short. However, he remains constructive on US equities by year-end. His favourite call is not on the S&P 500, but rather on gold, emerging markets, Europe, and commodities, where they are long. The rest of the world is ignoring the US recession.
His other big idea is that multipolarity is seen as a risk due to the higher probability of conflict, but a multipolar world actually has many playable investment theses: The world will be inflationary (but not dramatically). Technological innovation – and thus productivity – will surprise to the upside. Commodities will surge as capex spending remains buoyant. Investors should long capex / short producers, in every sub-sector. Emerging markets – particularly the geopolitical promiscuous ones – will win.
My take is that we began a new secular equity bull market in October 2022. The market is climbing a wall of worry, and there is indeed a lot to be worried about. The market is trying to resolve and price the uncertainty around: a) inflation and whether it will continue to fall, b) the economy’s ability to function without zero rates, c) that earnings can hold up given the most anticipated recession ever.
There are few that are bullish today.
2. What is the Future of Work? What if we had a AI co-pilot or digital assistant for every profession? Or if we had the capability to create digital clones of ourselves to pursue our creative projects? These prospects aren’t too far-fetched thanks to the latest developments in AI. What does the future of work actually look like?
My new favourite podcast is called Possible, which sketches out the brightest version of the future—and what it will take to get there. Most of all, it asks: what if, in the future, everything breaks humanity's way? It’s hosted by Reid Hoffman and Aria Finger. Each episode features an interview with a visionary from a different field: climate science, media, criminal justice, and more. The conversation also features another kind of guest: GPT-4, OpenAI’s latest and most powerful language model to date. Each episode has a companion story, generated by GPT-4, which will serve as a jumping-off point for a hopeful, speculative discussion about what humanity could possibly get right if we leverage technology—and our collective effort—effectively.
This week, Reid and Aria sat down with Jaime Teevan, Chief Scientist and Technical Fellow at Microsoft, and former technical advisor to CEO Satya Nadella. Leading Microsoft’s Future of Work Initiative, Jaime explores how everything from AI to hybrid work changes the way people get things done. Together, Jaime and our hosts discuss how we might best use AI, get a lot done with microtasks, and reimagine measures of productivity. Plus, GPT-4 and special guest stars share some neat stats and hot takes on the future of work.
Loved it.
3. This was a great debate and went deep into the obesity epidemic in the US.
Ozempic, the brand name drug for a medication called semaglutide, is one of the most popular drugs on the market right now. Originally developed to treat type 2 diabetes, the injectable drug has recently boomed in popularity for its off-label use to help people lose weight.…fast. Celebrities and public figures have admitted they're taking it.
But alongside the rise in Ozempic prescriptions come many questions still unknown: Who should be taking it? Is it safe for longterm use? Who is it safe for? Should children be prescribed it to treat childhood obesity, as the American Academy of Pediatrics recently advised? Is Ozempic a permanent solution to the obesity epidemic?
Or is it more like a bandaid, a quick fix that does little to address the root causes of obesity? And, to that end, what is the root cause of obesity? Is it a "brain disease," as one Harvard doctor recently declared on 60 Minutes that warrants medication? Or do diet, exercise, willpower and other behavioral lifestyle choices still matter?
Bari Weiss on the Honestly podcast had Dr. Chika Anekwe (an obesity medicine physician at Massachusetts General Hospital and an instructor in medicine at Harvard Medical School), Dr. Vinay Prasad (a hematologist-oncologist and a professor at the University of California San Francisco), and Calley Means (a former consultant for food and Pharma companies who now works to expose their practices and instead incentive healthy food as the foundation of health policy) to discuss what is causing so many Americans to be obese, which then leads to so many other diseases including diabetes, heart disease, cancer….
4. My friend Lorin Gu manages Recharge Capital, his family’s investment vehicle. He’s got a unique angle on the world and really takes family office investment to a new level. He was on the Village Global podcast discussing their approach.
Every time I hang out with him I come back with my mind expanded and motivated to learn new things.
Firstly, Recharge structures its direct investing thematically, rather than by asset class. The three themes that they believe have multi-decade tailwinds behind them: semiconductors, women’s health, and fintech/crypto.
Lorin had some great lessons from his time with David Swensen, particularly around the importance of the qualitative measurement of the people running the fund - including some great questions he asks them.
He then spends a lot of time talking about their favourite trends and how they are approaching them.
They close off on how to invest in light of the techno-nationalism around the globe.
Great job Lorin.
B. The Technology Section:
1. If you are new to Large Language Models (LLMs), the recent Economist addition did a great job explaining LLMs in this article: Large, creative AI models will transform lives and labour markets.
Key bits here:
Despite that feeling of magic, an LLM is, in reality, a giant exercise in statistics. Prompt Chatgpt to finish the sentence: “The promise of large language models is that they…” and you will get an immediate response. How does it work?
First, the language of the query is converted from words, which neural networks cannot handle, into a representative set of numbers. GPT-3, which powered an earlier version of ChatGPT, does this by splitting text into chunks of characters, called tokens, which commonly occur together. These tokens can be words, like “love” or “are”, affixes, like “dis” or “ised”, and punctuation, like “?”. GPT-3’s dictionary contains details of 50,257 tokens.
GPT-3 is able to process a maximum of 2,048 tokens at a time, which is around the length of a long article in The Economist. GPT-4, by contrast, can handle inputs up to 32,000 tokens long—a novella. The more text the model can take in, the more context it can see, and the better its answers will be. There is a catch—the required computation rises non-linearly with the length of the input, meaning slightly longer inputs need much more computing power.
The tokens are then assigned the equivalent of definitions by placing them into a “meaning space” where words that have similar meanings are located in nearby areas.
The LLM then deploys its “attention network” to make connections between different parts of the prompt. Someone reading our prompt, “the promise of large language models is that they…”, would know how English grammar works and understand the concepts behind the words in the sentence. It would be obvious to them which words relate to each other—it is the model that is large, for example. An LLM, however, must learn these associations from scratch during its training phase—over billions of training runs, its attention network slowly encodes the structure of the language it sees as numbers (called “weights”) within its neural network. If it understands language at all, a LLM only does so in a statistical, rather than a grammatical, way. It is much more like an abacus than it is like a mind.
Once the prompt has been processed, the LLM initiates a response. At this point, for each of the tokens in the model’s vocabulary, the attention network has produced a probability of that token being the most appropriate one to use next in the sentence it is generating. The token with the highest probability score is not always the one chosen for the response—how the llm makes this choice depends on how creative the model has been told to be by its operators.
The LLM generates a word and then feeds the result back into itself. The first word is generated based on the prompt alone. The second word is generated by including the first word in the response, then the third word by including the first two generated words, and so on. This process—called autoregression—repeats until the LLM has finished
The recent success of LLM in generating convincing text, as well as their startling emergent abilities, is due to the coalescence of three things: gobsmacking quantities of data, algorithms capable of learning from them and the computational power to do so. The details of GPT-4’s construction and function are not yet public, but those of GPT-3 are, in a paper called “Language Models are Few-Shot Learners”, published in 2020 by OpenAI.
To train, the LLM quizzes itself on the text it is given. It takes a chunk, covers up some words at the end, and tries to guess what might go there. Then the LLM uncovers the answer and compares it to its guess. Because the answers are in the data itself, these models can be trained in a “self-supervised” manner on massive datasets without requiring human labellers.
The model’s goal is to make its guesses as good as possible by making as few errors as possible. Not all errors are equal, though. If the original text is “I love ice cream”, guessing “I love ice hockey” is better than “I love ice are”. How bad a guess is is turned into a number called the loss. After a few guesses, the loss is sent back into the neural network and used to nudge the weights in a direction that will produce better answers.
If you’d like to go deeper into LLMs, then the best no fluff, high signal piece is Stephen Wolfram on What Is ChatGPT Doing … and Why Does It Work?
Key bit here:
The basic concept of ChatGPT is at some level rather simple. Start from a huge sample of human-created text from the web, books, etc. Then train a neural net to generate text that’s “like this”. And in particular, make it able to start from a “prompt” and then continue with text that’s “like what it’s been trained with”.
The specific engineering of ChatGPT has made it quite compelling. But ultimately (at least until it can use outside tools) ChatGPT is “merely” pulling out some “coherent thread of text” from the “statistics of conventional wisdom” that it’s accumulated. But it’s amazing how human-like the results are. And as I’ve discussed, this suggests something that’s at least scientifically very important: that human language (and the patterns of thinking behind it) are somehow simpler and more “law like” in their structure than we thought. ChatGPT has implicitly discovered it. But we can potentially explicitly expose it, with semantic grammar, computational language, etc.
What ChatGPT does in generating text is very impressive—and the results are usually very much like what we humans would produce. So does this mean ChatGPT is working like a brain? Its underlying artificial-neural-net structure was ultimately modeled on an idealization of the brain. And it seems quite likely that when we humans generate language many aspects of what’s going on are quite similar.
When it comes to training (AKA learning) the different “hardware” of the brain and of current computers (as well as, perhaps, some undeveloped algorithmic ideas) forces ChatGPT to use a strategy that’s probably rather different (and in some ways much less efficient) than the brain. And there’s something else as well: unlike even in typical algorithmic computation, ChatGPT doesn’t internally “have loops” or “recompute on data”. And that inevitably limits its computational capability—even with respect to current computers, but definitely with respect to the brain.
It’s not clear how to “fix that” and still maintain the ability to train the system with reasonable efficiency. But to do so will presumably allow a future ChatGPT to do even more “brain-like things”. Of course, there are plenty of things that brains don’t do so well—particularly involving what amount to irreducible computations. And for these both brains and things like ChatGPT have to seek “outside tools”—like Wolfram Language.
But for now it’s exciting to see what ChatGPT has already been able to do. At some level it’s a great example of the fundamental scientific fact that large numbers of simple computational elements can do remarkable and unexpected things. But it also provides perhaps the best impetus we’ve had in two thousand years to understand better just what the fundamental character and principles might be of that central feature of the human condition that is human language and the processes of thinking behind it.
2. There is no AI by Jaron Lanier is compelling from a number of angles.
Here are some key bits from the essay in the New Yorker, which I recommend reading in full:
If the new tech isn’t true artificial intelligence, then what is it? In my view, the most accurate way to understand what we are building today is as an innovative form of social collaboration.
A program like OpenAI’s GPT-4, which can write sentences to order, is something like a version of Wikipedia that includes much more data, mashed together using statistics. Programs that create images to order are something like a version of online image search, but with a system for combining the pictures. In both cases, it’s people who have written the text and furnished the images. The new programs mash up work done by human minds. What’s innovative is that the mashup process has become guided and constrained, so that the results are usable and often striking. This is a significant achievement and worth celebrating—but it can be thought of as illuminating previously hidden concordances between human creations, rather than as the invention of a new mind.
We need to get better at saying what is going on inside them and why. This won’t be easy. The problem is that the large-model A.I. systems we are talking about aren’t made of explicit ideas. There is no definite representation of what the system “wants,” no label for when it is doing a particular thing, like manipulating a person. There is only a giant ocean of jello—a vast mathematical mixing.
At the same time, it’s not true that the interior of a big model has to be a trackless wilderness. We may not know what an “idea” is from a formal, computational point of view, but there could be tracks made not of ideas but of people. At some point in the past, a real person created an illustration that was input as data into the model, and, in combination with contributions from other people, this was transformed into a fresh image. Big-model A.I. is made of people—and the way to open the black box is to reveal them.
In a world with data dignity, digital stuff would typically be connected with the humans who want to be known for having made it. In some versions of the idea, people could get paid for what they create, even when it is filtered and recombined through big models, and tech hubs would earn fees for facilitating things that people want to do.
A data-dignity approach would trace the most unique and influential contributors when a big model provides a valuable output. For instance, if you ask a model for “an animated movie of my kids in an oil-painting world of talking cats on an adventure,” then certain key oil painters, cat portraitists, voice actors, and writers—or their estates—might be calculated to have been uniquely essential to the creation of the new masterpiece. They would be acknowledged and motivated. They might even get paid.
Anything engineered—cars, bridges, buildings—can cause harm to people, and yet we have built a civilization on engineering. It’s by increasing and broadening human awareness, responsibility, and participation that we can make automation safe; conversely, if we treat our inventions as occult objects, we can hardly be good engineers. Seeing A.I. as a form of social collaboration is more actionable: it gives us access to the engine room, which is made of people.
3. My smart friend Bobby Yerramilli-Rao, now Chief Strategy Officer at Microsoft spoke about “How to Capture the AI Moment” in this podcast.
Lots of great stuff here:
What will be the final outcome of model structure & infrastructure? Who are the winners?
Where are key areas of value capture: Infrastructure where LLMs run, the LLMs themselves, the Platforms using LLMs, or at specific application layers?
Bobby spoke about the impact these models can have on education, healthcare and sales. The key to using them will be augmentation of current workflow, rather than creating something new.
This Microsoft video of their co-pilot is a good window into the Future of Work with AI.
4. Ilya Sutskever, chief scientist of Open AI and Jensen Huang, CEO of Nvidia. Two of the most consequential individuals in the proliferation of AI sit down together and talk about the present and future of AI.
5. Promptly learn prompting. Microsoft published a detailed guide to prompt engineering techniques. It includes system messages to prime the model with useful context, few-shot learning to teach new tasks, and priming the output.
Learning how to prompt is absolutely crucial for every person. First, AI could become our interface with the internet, and perhaps much of the physical world. Second, most AI will rely on the “human in the loop” to provide instructions. Prompting an AI with skill is how you get the most out of the technology, and do so safely.
6. A16Z shared their 2023 Crypto report. Three useful pages:
C. Prisoners of Geography
Got around to reading Prisoners of Geography: Ten Maps That Tell You Everything You Need to Know About Global Politics by Tim Marshall.
The book was a New York Times Best Seller and #1 Sunday Times bestseller. It covers ten regions or countries that have significant geopolitical roles or challenges: Russia, China, the United States, Europe, the Arab World, South Asia (mainly India and Pakistan), Africa, Japan and Korea, Latin America, and the Arctic. The book explains how geography has shaped their past and present, and what it means for their future.
One of the coolest parts of the book is the quotes he uses to describe different countries, my favourites:
China: “China is a civilisation pretending to be a nation” - Lucian Pye
Europe: “Here the past was everywhere, an entire continent sown with memories” - Miranda Mouillot
India: “India is not a nation, nor a country. It is a subcontinent of nationalities” - Muhammad Ali Jinnah.
The key ideas in the book:
Geography is a major factor in shaping the history and current affairs of different countries and regions in the world. Geography influences the opportunities and challenges that nations face, such as their access to resources, trade, security, and influence.
Geography is not destiny, but it does limit or enable the choices and actions of states and leaders. Geography can also interact with other factors, such as culture, ideology, religion, and technology, to create complex and dynamic geopolitical situations.
Geography can help us understand the motivations and interests of various actors in the world, as well as the sources of conflict and cooperation among them. Geography can also help us anticipate the possible scenarios and outcomes of global events and trends.
It reminded me a lot of Peter Zeihan’s Disunited Nations (which I covered in 2020). Both books are about how geography shapes the politics and power of different countries and regions in the world, but they have different perspectives and arguments.
Prisoners of Geography focuses on how physical features such as mountains, rivers, oceans, and climate limit or enable the choices and actions of various states. He argues that geography is a major factor in explaining the historical and current patterns of conflict and cooperation among nations.
While Disunited Nations focuses on how the decline of American leadership and the breakdown of the global order will affect the future of different countries and regions. He argues that geography is a major factor in determining which countries will thrive or struggle in a more chaotic and competitive world.
The biggest criticism of the books would be that they overstate the role of geography and ignore other factors such as culture, ideology, institutions, and human agency. They also tend to make sweeping generalisations and predictions that may not account for the complexity and uncertainty of global politics.
Related to the question of geography is the role of the climate and environment in human history. To investigate this, I’ve begun reading Peter Frankopan’s The Earth Transformed. More on that in a few weeks.
Time assets vs. Time debts by James Clear
Time assets are choices that save you time in the future. Think: saying no to a meeting, automating a task, working on something that persists and compounds.
Time debts are choices that must be repaid and cost you time in the future. Think: saying yes to a meeting, doing sloppy work that will need to be revised, etc.
Time assets are an investment. Time debts are an expense.
If you are in London, go check out Frameless. It’s awesomely beautiful.
Believe it or not, that “♡ Like” button is a big deal – it serves as a proxy to new visitors of this publication’s value. If you enjoyed this, don’t be shy.
A Few Things: Marko on Markets, The Future of Work, Solving American Obesity, Family Office Thematic Investing, LLM 101, There is No AI, How To Capture the AI Moment, A16Z on Crypto.....
Pound for pound one of the best newsletters out there